henhaobio
Sign in

bioRxivGeneticsPreprint

Complex-Traits Genetics Virtual Lab: A community-driven web platform for post-GWAS analyses

Gabriel Cuéllar-PartidaiD, Mischa LundbergiD, Pik Fang KhoiD, Shannon D’UrsoiD, Luis F. Gutiérrez-MondragóniD, Liang‐Dar HwangiD

PDFDOI
In-depth full-text digest available · Jump to digest ↓

Abstract

BackgroundGenome-wide association studies (GWAS) are an important method for mapping genetic variation underlying complex traits and diseases. Tools to visualize, annotate and analyse results from these studies can be used to generate hypotheses about the molecular mechanisms underlying the associations. FindingsThe Complex-Traits Genetics Virtual Lab (CTG-VL) integrates over a thousand publicly-available GWAS summary statistics, a suite of analysis tools, visualization functions and diverse data sets for genomic annotations. CTG-VL also makes available results from gene, pathway and tissue-based analyses from over 1,500 complex-traits allowing to assess pleiotropy not only at the genetic variant level but also at the gene, pathway and tissue levels. In this manuscript, we showcase the platform by analysing GWAS summary statistics of mood swings derived from UK Biobank. Using analysis tools in CTG-VL we highlight hippocampus as a potential tissue involved in mood swings, and that pathways including neuron apoptotic process may underlie the genetic associations. Further, we report a negative genetic correlation with educational attainment rG = -0.41 {+/-} 0.018 and a potential causal effect of BMI on mood swings OR = 1.01 (95% CI = 1.00-1.02). Using CTG-VLs database, we show that pathways and tissues associated with mood swings are also associated with neurological traits including reaction time and neuroticism, as well as traits such age at menopause and age at first live birth. ConclusionsCTG-VL is a platform with the most complete set of tools to carry out post-GWAS analyses. The CTG-VL is freely available at https://genoma.io as an online web application.

In-depth summary

Claude Opus 5.5 · xhigh effort · 17 Oct 2026 · from the full text of version 4 (9 May 2019)

A web platform bundling LD-score regression, DEPICT, MetaXcan, SMR, GSMR and fastBAT over uploaded or 1,500-plus public GWAS summary statistics, with precomputed gene, pathway and tissue results.

Post-GWAS work usually means chaining separate web tools whose functions are largely segregated, so groups analysing the same GWAS do not necessarily use the same summary statistics or annotation data. CTG-VL puts the common steps behind one interface and adds a catalogue of precomputed gene, pathway and tissue results for more than 1,500 traits, which the authors say no other web platform offered at the time.

Key results

  1. ShownRe-analysing a back pain GWAS inside the platform, a gene-based test flagged 59 genes, 20 of them across 10 loci outside the signals reported in the source study, including NTRK1 and DIS3L2.

    fastBAT tested 24,097 genes against back pain summary statistics (N = 509,070), giving a Bonferroni threshold of P < 2.0e-6. The authors note NTRK1 matters for the development of pain-mediating sensory neurons and DIS3L2 has been linked to abdominal pain. The gene list sits in a supplementary table absent from this text. · section 2; Case study; Discussion

  2. ShownLD-score regression run through the platform reproduced the genetic correlations previously published for the same back pain data, including the null correlation with height.

    Genetic correlation (rg) with BMI 0.34 +/- 0.02 (P = 1.02e-63), smoking 0.23 +/- 0.02 (P = 2.05e-16), neck/shoulder pain 0.72 +/- 0.11 (P = 8.18e-11), educational attainment -0.45 +/- 0.02 (P = 2.48e-105) and height 0.01 +/- 0.02 (P = 0.68), in line with the source study's findings. · Case study

  3. ShownTissue enrichment depended on how input variants were picked: central nervous system regions surfaced only with independent SNPs taken at P < 1e-5, nothing was significant at the genome-wide threshold, and no pathway passed either way.

    DEPICT was run at both thresholds used in the source study (P < 1e-5 and P < 5e-8) with FDR < 5%. CTG-VL selects independent SNPs by clumping within a 500 kb window and uses the European 1000 Genomes panel, whereas the source study selected them by conditional analysis against a reference of 10,000 UK Biobank individuals; the authors offer both differences as possible reasons the tissue results diverge. · section 2; Case study

  4. SuggestedTwo-sample Mendelian randomisation in the platform pointed to a causal effect of height on back pain but not BMI, with a heterogeneity test that leaves pleiotropic instruments open.

    GSMR using GIANT consortium exposures chosen to exclude UK Biobank: height OR = 1.06 per cm (95% CI 1.04-1.08) and BMI OR = 1.04 per kg/m2 (95% CI 0.99-1.09). The HEIDI tests were marginal (P = 0.057 for height, P = 0.0504 for BMI), consistent with some instrument SNPs affecting the outcome other than through the exposure. · Case study

  5. ShownThe precomputed catalogue was built from 3,670 Neale Lab GWAS, of which 1,747 with nominally significant heritability went through DEPICT and MetaXcan; querying it linked back pain pathways to several behavioural traits.

    Of 17 pathways suggestively associated with back pain (P < 0.0005), 9 were also associated with self-reported tension ('highly strung'), 8 with walking as a mode of transport and 6 with mood swings. MetaXcan used whole-blood prediction models only; DEPICT input came from clumping with a 500 kb window, R2 < 0.05 and P < 1e-5, relaxed to 1e-4 for traits with fewer than 10 independent loci. · Case study; Post-GWAS analyses in >1500 traits

Methods, data and reproducibility

A web application (Angular 6 interface, Node.js server) wraps PLINK clumping, LD-score regression, DEPICT, MetaXcan, SMR, GSMR and fastBAT around user-uploaded or hosted GWAS summary statistics, with 1000 Genomes phase 3 LD references, GTEx eQTLs and ENCODE/Roadmap chromatin annotations on GRCh37, plus interactive Manhattan and LocusTrack plots.

  • GWAS summary statistics hosted in the platform, with post-GWAS results precomputed using DEPICT and MetaXcan · human · n = more than 1,500 traits
  • Neale Lab GWAS summary statistics downloaded and variant-filtered, run through LD-score regression; those with heritability P < 0.05 taken forward · human · n = 3,670 downloaded, 1,747 selected
  • Back pain GWAS meta-analysis summary statistics used for the case study · human · n = N = 509,070
  • Height and BMI GWAS from the GIANT consortium without UK Biobank samples, used as Mendelian randomisation exposures · human
  • Reference data: 1000 Genomes phase 3 genotypes for LD, GTEx eQTLs (48 cell types for SMR), ENCODE and Roadmap chromatin states, RefSeq and UniProt gene descriptions · human
Methods & tools
CTG-VL (release 0.39-alpha), PLINK, LD-score regression with munge.py, DEPICT, MetaXcan with PredictDB models, SMR, GSMR (GCTA), fastBAT (GCTA), Tabix, NetworkX 1.8.1, Igraph 0.7.1, d3.js v5
Reagents & environment
Angular 6 interface with a Node.js server; network analysis scripts in Python 2.7 using NetworkX 1.8.1 and Igraph 0.7.1; Tabix for indexing user data; d3.js v5 graphics; platform release 0.39-alpha. The exact command for each analysis is displayed inside the platform.
Reproducibility gaps
The platform address and home page link are missing from the extracted text, so no URL can be listed, and no source repository is named. The release is alpha and updated weekly, so the live version may differ from the one described.
Limitations (6)
  • The report describes release 0.39-alpha and the platform is updated weekly, so the tools and data listed may not match what a user meets online. · stated by the authors
  • The first release ran MetaXcan only with whole-blood expression prediction models, with other tissues, plus broader SMR and fastBAT results, deferred to later updates. · stated by the authors
  • DEPICT tissue results differ from the published analysis of the same data, attributed to how independent SNPs are chosen (clumping versus conditional analysis) and to different LD reference panels. · stated by the authors
  • GSMR and SMR require exposure and outcome statistics from independent samples, and the HEIDI tests in the case study were marginal, so pleiotropic instruments cannot be ruled out. · stated by the authors
  • Validation is one case study against one published analysis rather than a systematic benchmark; the comparison with FUMA-GWAS, LD-Hub and MR-Base is a feature table, and the table itself is missing from this text. · noted in this summary
  • The precomputed catalogue draws on a single source of GWAS and a lenient heritability filter (P < 0.05), so some of the 1,747 traits carried forward may be weakly powered. · noted in this summary

Read this if you run post-GWAS workflows such as LD-score regression, DEPICT, MetaXcan, SMR, GSMR or fastBAT and want to know what one hosted platform covers and how its defaults shift the answers.

Citations
108

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISPasaniuc '16Mancuso '16Li '23Porcu '19Gaspar '18Zietz '26
59% semantic matchGenetics
Bogdan Pasaniuc, Alkes L Price·bioRxiv·Sep 1, 2016